Industrial Process Transfer Learning for Multi-Signal Model Reuse
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for industrial plants require tedious and time-consuming training and validation for each new problem, making it inefficient to reuse prior learning and knowledge across similar industrial applications, especially when dealing with multiple signals related to a single process.
Innovation Solution
A method of transfer learning that involves providing data templates for expected data, determining a process instance to map plant data, using historic process data to create a structured data matrix, and training a new machine learning model using a pre-trained model, allowing for the reuse of weights and layers while adjusting them according to new data, with pre-processing steps to standardize the data format.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a new machine learning model is trained for each new industrial plant problem, then the model can be accurately tailored to the specific plant's data and processes, but the training and validation process becomes tedious and time-consuming
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models on data from reference industrial plants before deployment. The pre-trained models contain prior knowledge and patterns learned from existing plant operations, which are then transferred to target plants. This eliminates the need for training from scratch, significantly reducing training time while maintaining model accuracy through transfer learning.
Solution Approach 2:
The patent uses copying by creating machine learning models based on pre-trained models from reference plants. Instead of developing entirely new models for each target plant, the system copies and adapts existing models, reusing their learned parameters and structures. This copying approach preserves accurate patterns while reducing the computational burden of training new models.
2Productivity
If machine learning models are reused across similar industrial plants, then training effort and cost are reduced, but the complexity of organizing and mapping input signals increases
Solution Approach 1:
The patent applies universality by creating a standardized signal mapping framework that can be applied across multiple industrial plants. The system defines universal signal categories and mapping rules that work for different plant types and processes. This universal approach allows the same pre-trained models to be transferred to various target plants by simply configuring signal mappings, rather than creating plant-specific model structures.
Solution Approach 2:
The patent uses segmentation by dividing the signal mapping process into discrete, manageable components. The system segments signals into categories (process signals, quality signals, equipment signals) and establishes mapping rules for each category. This segmentation simplifies the organization of input signals by breaking down the complex mapping task into structured, reusable mapping templates that can be systematically applied.
3Loss of information
If pre-trained models are transferred to new industrial plants, then prior knowledge is reused effectively, but the different signal spaces between similar tasks and plants make the transfer non-trivial
Solution Approach 1:
The patent applies parameter changes by transforming and normalizing signals from different plant signal spaces into a unified representation. The system adjusts signal parameters (scaling, normalization, feature extraction) to make them compatible with the pre-trained model's expected input space. This parameter transformation enables the transfer of learned knowledge across different signal spaces while maintaining the model's ability to process diverse industrial data.
Data Source
AI summary
A method of transfer learning for a specific production process of an industrial plant includes providing data templates defining expected data for a production process, and providing plant data, wherein the data templates define groupings for the expected data according to their relation in the industrial plant; determining a process instance and defining a mapping with the plant data; determining historic process data; determining training data using the determined process instance and the determined historic process data, wherein the training data comprises a structured data matrix, wherein columns of the data matrix represent the sensor data that are grouped in accordance with the data template and wherein rows of the data matrix represent timestamps of obtaining the sensor data; providing a pre-trained machine learning model using the determined process instance; and training a new machine learning model using the provided pre-trained model and the determined training data.